SQL服务器：3个月的分区怎么办？

Question

我想统计0-3个月分区的行数。月份由 MYMONTH 指定，格式为 201601 表示 2016 年 1 月。我正在使用 SQL Server 2014。如何在 3 个月内完成分区？

SELECT  COUNT(*), 
        COUNT(*)
           /  
        (COUNT(*) OVER (PARTITION 
            BY MYMONTH RANGE BETWEEN  3 MONTH PRECEDING AND CURRENT MONTH))
FROM myData

示例

|  Month  | Value  |  ID |
-------------------------|
| 201601  |   1    |  X  |
| 201601  |   1    |  Y  |
| 201601  |   1    |  Y  |
| 201602  |   1    |  Z  |
| 201603  |   1    |  A  |
| 201604  |   1    |  B  |
| 201605  |   1    |  C  |
| 201607  |   1    |  E  |
| 201607  |   10   |  EE |
| 201607  |   100  |  EEE|

计数

| Month | Count | Count3M | Count/Count3M |
-------------------------------------------
| 201601|   3   |    3    |     3/3       |
| 201602|   1   |    4    |     1/4       |
| 201603|   1   |    5    |     1/5       |
| 201604|   1   |    6    |     1/6       |
| 201605|   1   |    4    |     1/4       |
| 201607|   3   |    5    |     3/5       |

Answer 1

如果要统计前三个月的行数，使用条件聚合即可。您确实需要一种枚举月份的方法：

SELECT COUNT(*), 
       SUM(CASE WHEN yyyymm_counter <= 3 THEN 1 ELSE 0 END) 
FROM (SELECT md.*,
             DENSE_RANK() OVER (ORDER BY MYMONTH DESC) as yyyymm_counter
      FROM myData md
     ) md;

没有子查询的另一种方法是将月份值转换为实际日期。让我假设它是一个字符串：

SELECT COUNT(*),
       SUM( CASE WHEN DATEDIFF(month, CAST(MYMONTH + '01' as DATE), GETDATE()) <= 3
                 THEN 1 ELSE 0
            END)
FROM MyData;

我在答案中遗漏了 /。您需要注意 SQL 服务器进行整数除法，因此您可能得不到想要的结果——除非您将值转换为非整数（我建议乘以 1.0 或使用 1.0 而不是查询中的 1。

Answer 2

您可以试试这个 (MSSQL 2012)：

示例数据

CREATE TABLE mytable(
   MONT INTEGER  NOT NULL  
  ,Value INTEGER  NOT NULL
  ,ID    VARCHAR(5) NOT NULL
);
INSERT INTO mytable(MONT,Value,ID) VALUES (201601,1,'X');
INSERT INTO mytable(MONT,Value,ID) VALUES (201601,1,'Y');
INSERT INTO mytable(MONT,Value,ID) VALUES (201601,1,'Y');
INSERT INTO mytable(MONT,Value,ID) VALUES (201602,1,'Z');
INSERT INTO mytable(MONT,Value,ID) VALUES (201603,1,'A');
INSERT INTO mytable(MONT,Value,ID) VALUES (201604,1,'B');
INSERT INTO mytable(MONT,Value,ID) VALUES (201605,1,'C');
INSERT INTO mytable(MONT,Value,ID) VALUES (201607,1,'E');
INSERT INTO mytable(MONT,Value,ID) VALUES (201607,10,'EE');
INSERT INTO mytable(MONT,Value,ID) VALUES (201607,100,'EEE');

查询 1

SELECT MONT, RC,  RC+ LAG(RC,3,0)  OVER (  ORDER BY MONT)+ LAG(RC,2,0)  OVER (  ORDER BY MONT)  + LAG(RC,1,0)  OVER (  ORDER BY MONT)   AS RC_3M_PREC -- + COALESCE( LEAD(RC)  OVER (  ORDER BY MONT),0) AS RC_3M
FROM (SELECT MONT
    , COUNT(*) RC
     FROM mytable
     GROUP BY MONT
 ) A

输出：

MONT        RC          RC_3M_PREC
----------- ----------- -----------
201601      3           3
201602      1           4
201603      1           5
201604      1           6
201605      1           4
201607      3           6

或使用您建议的内容（选项 ROWS ... PRECEDING）：

查询 2：

SELECT MONT, RC
    ,   COALESCE(SUM(RC)   OVER (ORDER BY MONT ROWS BETWEEN 3 PRECEDING AND CURRENT ROW),0) AS RC_3M
FROM (SELECT MONT
    , COUNT(*) RC    
FROM mytable
GROUP BY MONT
 ) A

输出：

MONT        RC          RC_3M
----------- ----------- -----------
201601      3           3
201602      1           4
201603      1           5
201604      1           6
201605      1           4
201607      3           6

SQL服务器：3个月的分区怎么办？

SQL Server: how to do 3 months' partition?

sql

sql-server

time

partition